SmartTrans

Official implementation of SmartTrans: Advanced Similarity Analysis for Detecting Vulnerabilities in Ethereum Smart Contracts. The Model and Dataset are released at NextCloud

Prerequisites

• Linux OS (Tested on Ubuntu 24.04.2 LTS)
• Python 3.8+
• PyTorch 1.10+ with CUDA 10.2+
• NVIDIA Drivers (Tested with 550.135)
• Gigahorse-toolchain (for bytecode CFG analysis)

Quick Start

Environment Setup

1. Create and activate conda environment:

conda create -n smarttrans python=3.8 -y
conda activate smarttrans

2. Install dependencies:

pip install tqdm pandas transformers faiss-gpu accelerate datasets wandb

3. Clone repository:

git clone https://github.com/vul337/SmartTrans.git && cd SmartTrans

4. Download and extract models and datasets: Please download and extract the models from NextCloud

tar -xzvf *.tar.gz
mkdir index/
mv searching_index.index index/
mv searching_ebds_list.pkl index/

Usage

Analyzing Custom Smart Contracts

1. Preprocess bytecode using Gigahorse-toolchain (set TEMP_PATH in 1.Analyze_gigahorse.py to point to Gigahorse's .temp directory)

python3 1.Analyze_gigahorse.py

2. Run preprocessing and inference:

python3 2.Preprocess.py
python3 3.SaveEmbedding.py

3. Retrieve similar functions:

python3 4.Retrieve.py

Training

Pre-Training

train.sh

Finetuning

finetune.sh

Reproducibility

The exact conda environment configuration is provided in smarttrans_env.yaml.

Citation

If this work is helpful for your research, please consider citing the following BibTeX entry.

@article{chen2025smarttrans,
  title={SmartTrans: Advanced Similarity Analysis for Detecting Vulnerabilities in Ethereum Smart Contracts},
  author={Chen, Longfei and Wang, Hao and Zhou, Yuchen and Wong, Taiyu and Wang, Jialai and Zhang, Chao},
  journal={IEEE Transactions on Dependable and Secure Computing},
  year={2025},
  publisher={IEEE}
}